The Cold Start Problem: How to Start and Scale Network Effects by Andrew Chen Part V — The Ceiling (Book Notes)
Chapter 22 — Twitch
Hitting the ceiling is painful, especially after years of endless growth. Exponential growth curves can turn into a squiggle? Why. Because there are negative forces that appear during the last stage of the lifecycle of a network. Market saturation, churn for early users, bad behaviors from trolls, spammers, and fraudsters. Lower quality engagement from new users, regulatory action, degraded product experience as too many users join. When users are leaving a network as fast as new users are signing up the top-line growth naturally slows.
Growth curves of even the top products grow in fits and starts. When a ceiling is hit product teams scramble to ship the right features to push off the ceiling. Network effects can unravel just as fast as they gather.
Twitch had grown to millions of users but hit a ceiling. Earlier on in 2010, the original idea of the product was to focus on streaming content of all types, not just gaming. It had grown well but plateaued. The company was profitable, but they hit an impasse where they weren’t growing at all.
Justin from JustinTV was the first streamer on the platform and many tech insiders used to tune in. This created the first atomic network. They would eventually let others stream as well. This grew to people streaming singing, dancing, sports broadcasts and of course video games. At JustTV growth eventually stalled at a few million users and was stagnant.
To break through this ceiling they had to scale the product and take a bigger bet. They decided to go after a few opportunities all at once. The first part of the team would work on mobile video and streaming. Second, a small team would focus on video games.
They did a lot different with Twitch than JustinTV. They worked on tools for streamers that the team improved over time. Making money was important to streamers to they added tipping features. They also redesigned the whole website to make streamers discoverable based on which game they were playing, sorted in a way that rewarded the most popular streamers. By focusing on games they could make all the changes in a way that served the streamers and audiences much better.
Many of the changes were jsut tweaks to JustinTV’s features but they would specifically benefit streamers. For example. Many of them wanted to push out high definition streams of their gameplay rather than low res versions that were common during that time. Hi-res helped viewers follow the action. Streamers can be more easily found if they could be found by the game they were playing. Adding categories for league of legends, pub G, GTA, and other popular games at the time. Twitch also decided to list streamers by the number of people watching so the most popular streamers would be at the top. A new partnership service was had to provide white-glove service to top streamers and up and comers. Twitch began to participate in large esports tournaments like the league of legends events. Then they started TwitchCon which brought viewers and streamers together.
The new features and functionality was built with the idea that the atomic network for twitch could be as small as one streamer and one viewer watching them. Playing videos games with even one twitch viewer is way more fun than playing by yourself. If they're watching and chatting while you play. There’s a human connection that you want to come back for. As the economic angle is added the depth of the streaming because more important the real magic starts to happen once you have enough followers on twitch and you consistently have viewers every time you start streaming. Then every session on twitch becomes fun since there's always an audience. Streamers talked about how even making $20 or $50 a month was an eye-opener.
If you can build enough of an audience you can go pro and make a living. Top streamers tended to make 300k+ a year. It's all about the streamers. It’s all about helping streamers create content, find audiences, and monetize. The combined strategy of new features emphasizing games content and addressing the needs of streamers worked. within a month after launch twitch had 8 million unique viewers within a year after that it had doubled to 20 million and doubled again. So much so that today twitch is one of the most trafficked websites in the world.
Even Facebook had hit this problem. Growth had stalled around 90 million. They hit a wall. Then they built their first growth team and built a series of products that helped them breakthrough. Getting user-profiles better indexed by google via SEO, creating recommendations for people you should add as friends.
Chapter 23 — RocketShip Growth T2D3
In the U.S. roughly 6 million businesses are started each year. Only a small number are suited to be Venture Capital investments. Only 1 in 20 startups end up with the 10x growth that the industry is focused on. 100’s of exits per year but only a few dozen are large enough to define the industry. In other words, even once a team is shown promising enough to be backed by investors. Very few ever make it to the other side in the form of an exit. Generally, the outcome is the same. Stop growing exit or die.
Setting a goal of exceeding a goal of a billion on valuation. Requires at least 100 million dollars in top-line ARR. Would want to hit that in 7–10 years. Revenue and time work together to create an over constraint. Saas Companies need to follow a precise path to reach these numbers. First, get to PMF then get to 2 million in ARR. Then Triple to 6 million. Then triple again to 18 million dollars of recurring revenue the double to 36. then double to 72 million then double one last time to exceed 100 and land on 144 Million. Saas companies like Marketo, Netsuite, Workday, Salesforce, Zendesk, and others have all roughly followed this curve and the rough timing makes sense. The first phase in which the team gets to PMF is 1–3 years. Add on the time to reach the rest of the growth milestones and the whole thing might take 6–9 years.
Chapter 24 — Saturation- eBay
In 2000 eBay’s U.S. business failed to grow on a month-over-month basis. Without growth in the U.S., the entire business would stagnate. It’s tempting to just optimize the core business. Theres’ no way around maintaining a high growth rate besides continuing to innovate. eBay released “Buy It Now” and it paid off big. Launching Buy It Now was a large change that touched every transaction. With initial success, they doubled down on innovation to drive growth. they introduced stores on eBay which dramatically increased the number of products offered for sale on the platform. They expanded the menu of optional features that sellers could purchase to better highlight their listings on a site. They improved the post-transaction experience by improving the checkout flow and then adding the seamless integration of PayPal on the eBay site. Each of these innovations helped support the business and kept stalling the business at eBay.
Uber’s impressive growth trajectory was a combination of launching in more and more cities each year while simultaneously leayering on new products like carpooling and food delivery.
Network Saturation Vs. Market Saturation
The 100th connection for every participant is often less impactful than the first few. As the network gets denser over time its associated network effects become less incrementally powerful. In eBay’s case when you search for something like vintage Rolex Daytona the product experience and associate conversation rate improve dramatically as you add the first few listings. It might even continue to improve with a few dozen. You don’t need the search to return 5,000 listings. The buyer is unlikely to browse through that many. Same with Uber adding the first 100 cars on the road is important. But diminishing returns arrive at large numbers.
Both Network Saturation and Market Saturation can slow down growth. You eventually run out of companies that can sign up for your collaboration tools. Or gamers for your massive multiplayer game.
How to fight these saturations is to constantly evolve your product, the target market, and the feature set. There’s no way around it.
New adjacent networks in the network of networks some will inevitably be more engaged than others. Instagram’s success was anchored on the Adjacent User Theory. The adjacent users are aware of a product and possibly tried using it but are not able to successfully become engaged users. This is typical because the product experience has too many barriers to adoption for them. While IG had PMF for over 400 million people. They discovered new groups of billions of users who didn't understand Instagram and how it fits into their lives. Rather than focusing on the core network of power users. Instead, the approach was to focus on the adjacent set of users whose experience was sub-par.
There might be multiple sets of adjacent networks at each moment. Like IG not having great support for low-end Android apps. Or it might be because of the quality of the networks. When Bengali first started at IG the Adjacent network was women 35–45 years old in the US, had a FB account but didn’t see the value of Instagram. By the time I left the adjacent user was woman in Jakarta on an older 3G Android phone with a pre-paid mobile plan.
Teams need to continually evolve their offering to the next set of sellers or creators to their platform. For example, when Uber ran out of full-time limo drivers for the service, the next set of adjacent users were people who had never driven as a form of income. Eventually, this pool is exhausted as well. Serving each adjacent network is like adding a new layer. Doing this requires the team to address new markets.
Adding new geographies is another way to build up the layer cake. This is participially obvious for products that act on a hyper-regional level. Companies like Open Table, Yelp, and Uber. And Others that grow city to city. Each new region provides a fresh market to grow. Regional expansion is easier when the network grows into directly adjacent networks. When a hyper-local network in SF wants to expand to a nearby city like Los Angeles that often works well because both markets will share users. OpenTable might be able to leverage the fact that restaurant chains are regional. This makes it easy to launch in new markets.
Chapter 25 — The Law of shitty click-throughs — Banner Ads
Not really much here besides — Every marketing channels degrade over time. Regardless of whether you’re talking about email, paid marketing, social or video. This is a core reason why products hit growth ceilings. The first ads ever had incredible engagement 78% to start now, it’s a 100th of this percentage. Email marketing followed this trend too. Don’t send too many emails.
Chapter 26— When the Network Revolts — Uber
The Hard Thing about the hard side of the network. Ebay’s hard side is its sellers who’ve revolted many times whenever listing fees have changed, same with Airbnb hosts, Instacart workers, and Amazon sellers. A well-organized revolt can kill a product in its entirety. On Vine, creators were driving billions of views and threatened to leave the platform if certain aspects weren’t built out. Vine turned down the plan and a few years later the service was shut down. The hard side is worth the effort to cultivate. If they can be retained they can be the backbone of any product. Slack’s S1 showed that less than 1% of slack’s customers accounted for 40% of the revenue. Zoom’s 30% of revenue came from just 344 accounts. Again, less than 1% of their customer base. As the network scales its hard side will professionalize. Quality and consistency will increase and the best players will be able to do it at scale. There is no choice but to embrace this.
How professionalization happens.
Happens in 2 ways. Homegrown professionals and Off-network
You might start selling vintage clothing on the site to make money but find that you can do it full time yourself. After some time you may be able to start your own boutique and become a power seller. There are millions of businesses like this selling on eBay / Amazon and other eCommerce platforms. The B2B version might start out wanting to try out a new product. Having a specific team and a set of experts on it. Then having consultants and vendors professionally implement it for a broader ecosystem. That’s what has happened before for business software like CRM.
Largest off-network players joining overtime.
It was only years later than Microsoft put its Apps onto iOS. The CEO of Microsoft at the time wanted to support Microsoft on all platforms not just Microsoft ones. Nintendo held out for years, they only had Mario and Zelda only on their own hardware devices which they hoped would serve as anchors to their own network ecosystem. But they eventually released an iOS app once mobile was too big to ignore. When a netowrk becomes large rich and diverse it’s often described as an economy. Like the Gig economy, the attention economy, and creator economy. Each of these respectfully encompasses the worlds of Airbnb, Uber, and Instacart.
At times it makes more sense to scale up than to acquire more users to the hard side.
Chapter 27 Eternal September — Usenet
Before Snapchat, Facebook, or even Geocities or yahoo groups. Think of it as the very first social network. Created in 1980. It was the first worldwide distributed discussion system. Hosting newsgroups like talk.politics, rec.arts.movies, rec.crafts.wine.making +100 other topics. For the early internet, Usenet was a big deal. Important announcements were made there like the launch of the world wide web.
Since Usenet had the most people and the most comprehensive set of topics why would you participate in discussions anywhere else? But then by 2000 Usenet was practically dead. they hit the ceiling and never recovered. the problems that Usenet ran into are the same ones that plague social networks today.
Context collapse —
Every network if it starts with a focused atomic network has a shared context of what you should and shouldn’t do within the network. A culture.
The story is from Adam Diangelo, CEO of Quora and former CTO of Facebook.
When you first join a social network with your close friends it’s easy to use it a lot. You might post photos and comments all the time full of in-jokes and stories. You and your friends like it so much that you invite your other friends and siblings too. But eventually photos and content might attract comments from people you don’t know well. Parents and teachers. Those photos of a party you went to might get you in trouble. Context collapse is most problematic for social networks, because they can’t satisfy everyone in every context. How can creators ensure that they can properly convey their message? How will they know they won’t offend or get judged harshly if the wrong audience comes across it. If this happens creators are less likely to create which means they can lose a whole sub-community.
Networks of Networks… of Networks
So how do you prevent network collapse? Products like iMessage or whatsApp can give us a clue. Messaging apps are resistant to context collapse. Slack channels offer a different experience. As more and more people join they set up more spaces for close teammates. When users are able to group themselves networks prove to be resilient. Facebook groups and snap stories both provide a network within a network that provides its own context. Like Finstas. Product features like the time zone warning on Slack let you know that there is a different context.
Chapter 28 — Overcrowding — Youtube
When Youtube got to millions of videos it got hard to figure out exactly what to watch. Steve Chen — co-founder of Youtube said too many videos can make the product unusable. it’s what happens when there are too many emails, comments, and threads. When you follow too many people on your social media app and there’s too much to deal with.
Early days of organizing videos
Youtube as a dating site didn’t last long. After a few weeks, the founders realized it was a good idea to open it up to any video being uploaded. The first video uploaded after that was titled, “Me at the zoo”. In the earliest days there was little content to organize. Organizing the videos was an afterthought. They focused on key features that allowed the product to be shared and embedded basically solving the cold start problem. Once Youtube got a lot of videos on the platform they had to make it easier to discover the best videos. Youtube made a top 100 videos list. Then it made sense to make categories to further organize the content. After that Youtube added comments. In less than a year Youtube hit 1 million views per day.
For a marketplace startup. In the early days, the relatively limited choices means sellers don’t compete with each other avoiding overcrowding. Consumers have a more focused catalog from which to shop. once it grows to millions of users there might be hundreds of fo sellers for every product and picking the best one may not be easy.
The power of Data and algorithms
Youtube was trying to keep up with all the traffic. They weren’t trying to come up with a bunch of new features because there was so much traffic. The few feature updates that they did ship focused on search, relevance, and algorithmic recommendations. In other words, the core levers to solve the overcrowding issues which would otherwise make Youtube a confusing fragmented place. This is where Google’s expertise in handling massive amounts of Data came in handy for the following years. Search and related videos. These helped users quickly navigate to videos they cared about and because they were algorithmically driven it didn't require the company to manually curate the content.
Better matching between creators and viewers alleviated the overcrowding problem in a product that has more than a billion users.